During this year's Two Sessions, Zhou Hongyi, a member of the National Committee of the Chinese People's Political Consultative Conference and founder of 360 Group, focused on the safety and illusion of large models. He believed that large models represented by DeepSeek have achieved "national science and technology achievements", which not only promoted independent innovation of large model technology, but also promoted the coordinated development of upstream and downstream industries, allowing large models to accelerate their entry into hundreds of industries and become a key force in promoting China's scientific and technological progress, economic growth and national prosperity.
However, large models including DeepSeek have "illusions" and occasionally generate situations that seem unreasonable and inconsistent with the facts. "Illusion" is a double-edged sword. It is an inherent characteristic of large models and cannot be eliminated. It also determines its creativity and imagination. It is extremely important for the realization of AGI (artificial general intelligence). If the large model has no "illusion", imagination and creativity will be lost. And the more powerful the model, the more hallucinations it generally has.
Zhou Hongyi believes that in the field of scientific research, the "illusion" of large models can "imagine" the molecular structure of new drugs, the structure of proteins, etc., and guide the direction of new research. In fields such as medicine, law, and finance that require high accuracy, although "illusions" may be wrong, they can be corrected by using technical means, such as RAG (Retrieval Augmented Generation) technology, by comparing professional knowledge bases, real-time network information, etc.
Zhou Hongyi said that looking back on the past 30 years, my country's laws and regulations have incorporated the "safe harbor principle" to allow companies to dare to invest in innovation, which has brought about the prosperity of the Internet industry and provided technical, industrial and talent support for the later "Internet +" and today's "Artificial Intelligence +". The development of large models is related to the destiny of the country, and its creativity cannot be restricted due to excessive pursuit of accuracy. He suggested that we learn from the successful experience of the "safe harbor principle", implement flexible supervision on large models and related products and services, moderately accommodate the "illusion" of large models, avoid direct shutdowns for general mistakes caused by "illusions", give companies the opportunity to self-correct, promote bold innovation, let go of competition, and strive to allow more companies to "copy" the success of DeepSeek.
Zhou Hongyi once again focused on the transformation of the network security industry and suggested "solving the dilemma of the network security industry with security as a service."
Network security is an important part of national security. In the process of national digitalization and intelligent development, security challenges are increasing day by day. Zhou Hongyi believes that in building a national network security defense system, we cannot only focus on large enterprises and state agencies. Local enterprises and private enterprises are also critical. They are an important part of the supply chain, industrial chain and ecological chain.
However, the current network security industry is seriously involved in involution, and the traditional model is full of problems. It is not only difficult to achieve broad coverage of security capabilities, but also unable to help enterprises and other customers quickly build security capabilities, which invisibly exacerbates national network security risks.
On the one hand, the hardware box sales model is a "lose-lose" model, with high investment but poor results. Manufacturers have also experienced soaring costs due to diversified customization needs, and are struggling to cope with customization tasks, hindering market expansion. On the other hand, traditional security software emphasizes marketing and neglects delivery, which often does not meet customer needs. Software as a Service (SaaS) has become a global trend. SaaS service products are visible, experienceable, cheap, simple to deliver and easy to promote, and can help enterprises quickly form security capabilities. For example, the transformation of the cloud computing industry is a successful example. However, the network security industry still remains at the traditional system integration level and cannot keep up with the needs of the security situation.
To this end, Zhou Hongyi put forward two suggestions to promote the service-oriented transformation of network security. One is to follow the global trend and encourage a shift from purchasing hardware products to purchasing security services. Relevant departments can incorporate the deployment of security services into special subsidy projects, provide financial subsidies or tax incentives to companies that actively adopt security services, and guide the market to transform into a service-oriented one. Second, they can encourage the purchase of SaaS-based security products and use their advantages to help companies adapt to needs at low cost, evaluate effectiveness, improve network security protection levels, and lay a solid security foundation for the country's digital construction.
According to reports, 360 has been advocating the concept of SaaS services in recent years, launched SaaS network security solutions, and explored a set of low-cost, low-threshold, lightweight SaaS service models for small and medium-sized enterprises, and has successfully helped millions of customers complete digital transformation.